为了提高交通水利和深部煤矿工程中的TBM智能化施工和灾害预测的能力，提出了一种结合卷积神经网络（CNN）和长短时记忆神经网络（LSTM）优势的CNN-LSTM模型，依托“引汉济渭”工程，预测TBM隧道稳定段掘进参数和岩爆等级。此外，对TBM数据进行清洗和预处理，根据桩号将TBM数据、地质参数和岩爆等级匹配，基于灰色关联分析筛选出合理的预测指标，合理选择CNN-LSTM模型的超参数以获得较好的预测结果，并与其他模型的预测结果进行对比分析。研究结果表明：对于TBM隧道稳定阶段的掘进参数推力F、贯入度P和扭矩M，CNN-LSTM模型预测的平均绝对误差（MAE）、平均绝对百分比误差（MAPE）和均方根误差（RMSE）明显小于其他模型且较稳定。其中，扭矩M的预测效果最好，推力F的预测效果次之，贯入度P的预测效果最差；随着岩爆等级的增加，刀盘和刀具受到严重冲击，不同模型对于以上掘进参数的预测效果变差，但CNN-LSTM模型的预测结果均在有效范围内且具有较高的鲁棒性；CNN-LSTM模型预测TBM隧道岩爆等级的准确率（ACC）、精确率（PRE）和召回率（REC）明显高于其他模型，准确率（ACC）、宏观精确率（MPRE）和宏观召回率（MREC）分别达到98.17%、97.73%和98.58%；根据模型的随机采样分析可知，CNN-LSTM模型对于相同容量的不同岩爆样本的鲁棒性较高，明显优于其它模型。综上，CNN-LSTM模型对于TBM隧道的掘进参数及岩爆等级的预测具有可行性和有效性。

In order to improve the intelligent construction and disaster prediction capabilities of TBM in traffic water conservancy and deep coal mine engineering, the CNN-LSTM model combining the advantages of convolutional neural network (CNN) and long short-term memory neural network (LSTM) was proposed, and the tunnelling parameters in the stabilization stage of TBM tunnel and rockburst grade were predicted based on the Hanjiang River-Weihe River Water Conveyance Project. In addition, the TBM data was cleaned and preprocessed, the TBM data, geological parameters and rockburst grade were matched according to the station number, reasonable prediction indicators were screened out based on grey relation analysis. And the hyperparameters of the CNN-LSTM model were reasonably selected to obtain better prediction results, and the prediction results of other models were compared and analyzed. The research findings indicate that for the tunneling parameters of thrust (F), penetration (P), and torque (M) during the stable phase of TBM tunneling, the CNN-LSTM model predicts significantly lower and more stable mean absolute error (MAE), mean absolute percentage error (MAPE), and root mean square error (RMSE) compared to other models. Among them, the torque (M) prediction effect is the best, the thrust (F) prediction effect is second, and the penetration (P) prediction effect is the worst. With the increase of rockburst grade, the cutterhead and cutter are severely impacted, and the prediction effect of different models for the above tunnelling parameters become worse, but the prediction results of CNN-LSTM model are within the effective range and have high robustness. The CNN-LSTM model's accuracy (ACC), precision (PRE), and recall (REC) in predicting the rockburst grade of TBM tunnels are significantly higher than those of other models, with accuracy (ACC), macro-precision (MPRE), and macro-recall (MREC) reaching 98.17%, 97.73%, and 98.58% respectively. According to the random sampling analysis of the model, the CNN-LSTM model has high robustness for different rockburst sample of the same capacity, which is significantly better than other models. In conclusion, the CNN-LSTM model is feasible and effective for predicting the TBM tunnelling parameters and rockburst grade.